Reporting and BI in one paragraph
A reporting setup reads each platform through its own interface or a connector: Google Ads describes what Google Ads did, GA4Google's event-based analytics platform. describes what the website did, the CRM describes what sales did. Each is accurate about itself and silent about the others, and each keeps only as much history as it chooses to. Business intelligence copies the data out of every system into a store the team owns, keeps every day of it, joins spend to sessions to leads to revenue on shared keys, and publishes metrics whose definitions were agreed once and written down. The difference is not the chart at the end. It is whether the number behind the chart can be traced to its rows, joined to another system's rows and reproduced next month by a different person. The reporting versus BI comparison sets the two side by side; the reporting and BI overview covers the consumption side, meaning audiences, cadence and narrative. This section covers production: how trustworthy numbers get made.
The marketing BI stack
Every marketing BI setup, whatever the vendors, has the same layers. Teams differ in how many they build themselves and how many they buy, and in how far down the stack they have got.
| Layer | What it does | What marketing puts in it | Guide |
|---|---|---|---|
| Sources | The systems that generate data | Ad platforms, GA4, CRM, e-commerce platform, email tool, call tracking | Reporting vs BI |
| Ingestion | Copies source data into the warehouse on a schedule | Managed connectors, the GA4 BigQuery export, platform APIs, spreadsheet uploads for the rest | BigQuery export basics |
| Warehouse | Stores raw and modelled tables with full history | Daily spend, events, sessions, orders, customers, campaign metadata | SQL for marketers |
| Transformation | Cleans, standardises and joins raw tables into marts | Channel mapping, UTM parsing, currency normalisation, spend-to-outcome joins | SQL for marketers |
| Metrics layer | Holds each metric's definition once: formula, grain, filters, owner | Cost per lead, return on ad spend, blended acquisition cost, pipeline by source | Metrics layer and definitions |
| BI tool | Queries the layer and renders dashboards and explorations | Looker Studio, or a licensed BI product once the team outgrows it | Looker Studio section |
| Consumption | Dashboards, scheduled reports, ad hoc queries, alerts | Weekly performance views, monthly reviews, finance reconciliations | Dashboard governance |
Most marketing teams enter the stack from the bottom up: Looker Studio reading directly from connectors, then the GA4 BigQuery export as the first warehouse table, then spend tables beside it, then definitions once two dashboards disagree. That order is fine; the mistake is treating any one stage as finished. The arithmetic the warehouse models is the arithmetic of the KPI framework: outcomes decomposed into drivers, which is what a KPI tree writes down before anyone builds a table.
The core guides
- Reporting vs business intelligence — what each is for, where reporting is enough, what BI adds, and how teams move from one to the other without rebuilding everything at once.
- SQL for marketers — the handful of query shapes that answer most marketing questions: filtering and aggregating, time grains, joining spend to outcomes, window functions for cohorts, and the pitfalls that inflate totals.
- Metrics layer and definitions — why the same metric differs by tool, how to write a definition with formula, grain, filters and owner, and how to version and enforce it.
- Dashboard governance — ownership, certification, an inventory, naming and access rules, change management and the review that retires what nobody uses.
Two neighbouring sections do work this one depends on. The Looker Studio section covers the dashboard tool most marketing teams use first, including data sources, blending and performance. The attribution section covers the modelling question that BI projects usually end up being asked to answer, because once spend and outcomes sit in one table, someone will ask which channel deserves the credit (AttributionAssigning conversion credit to touchpoints.). Both also have a tutorial that works on warehouse tables: moving a Looker Studio report onto a warehouse and building an attribution model in SQL from the Google Analytics BigQuery export.
Further guides
Further guides go deeper into building the tables, reading what they show, checking and correcting what is shared, and running the team that uses them:
- Designing a marketing data model — the tables behind marketing reporting: facts for spend, sessions, leads and orders, shared dimensions for date, channel and campaign, and the keys that join them.
- Loading ad platform data into a warehouse — connectors or exports, refresh windows for late conversions, currency, time zones and renamed campaigns.
- Writing data tests for marketing tables — a tutorial for SQL tests that catch broken data before a dashboard does: duplicates, missing values, unmapped rows, freshnessWhether a table, feed or report holds data as recent as it should; a freshness check fails when the latest date is older than the source's expected arrival., volume and reconciliation.
- Explaining a change in a marketing metric — rule out data changes, split the metric into parts, separate mix from rate, find the segment and test the story.
- Comparing periods in marketing reports — weekdays and holidays aligned, partial periods, seasonality, tracking breaks and late conversions.
- Reading small numbers in marketing reports — why rates built on small counts jump around, ranges instead of single figures, minimum volumes, pooling and control limits.
- Running an analytics request process — intake that captures the decision, triage and sizing, self-serve first, reusable answers and closing the loop.
- Building data literacy in a marketing team — the skills that matter, definitions within reach, reading uncertainty, office hours and checks that it worked.
- Reviewing an analysis before it is shared — a tutorial for a second person's review of a one-off analysis: restate the question and the decision, reproduce the headline numbers, check definitions, joins and the counts behind every rate, pair each claim with its evidence and record the sign-off.
- Correcting a number after a report has gone out — a tutorial for a wrong figure in a sent report or live dashboard: confirm and size the error, fix the source, restate the figures with the old ones still findable, tell the people who acted on it first and add the test that would have caught it.
When a team needs BI
The honest answer for many teams is: not yet. A single channel, one platform and a reviewer who reads the platform's own reports is a reporting problem, and a warehouse would add cost without adding decisions. The signals that the balance has tipped are specific:
- The same metric differs between meetings. Two dashboards, two connectors, two definitions of a conversion. Nobody can say which is right because neither can be traced.
- Questions need joins. Cost per lead by landing page, revenue by first-touch channel, retention by acquisition campaign. Each needs rows from two systems on a shared key.
- History matters more than the platforms keep. Year-on-year comparisons, seasonality, cohorts that take a year to mature.
- Several people build their own reports, each from their own export, and the reports drift apart.
- Finance asks for reconciliation, and the marketing number has to tie to the invoice and the ledger.
- Attribution and incrementality questions cannot be answered from any one platform's view of itself.
When several of these are true at once, the cost of not having BI is already being paid in meetings, re-exports and wrong decisions. When none are, the right move is better reporting discipline: one dashboard, one owner, one definition of each KPIA metric tied directly to an objective..
A subscription box deciding whether to build BI
A subscription box business runs paid search, paid social and affiliate, with a CRM for subscribers and a spreadsheet for refunds. Monthly reporting takes the marketing lead two days: four exports, three pivot tables, one deck. In the latest review paid social shows 1,420 new subscribers at £38 each while finance counts 1,180 new paying subscribers across all channels, and nobody can reconcile the gap. The team costs the alternative: a managed connector for each ad platform, the GA4 export into BigQuery, a weekly SQL model that joins spend to subscriber records on the order id, and one Looker Studio dashboard reading from it. The build is estimated at six weeks of a part-time analyst and a monthly warehouse bill under £150. Against two days a month of the lead's time and a subscriber count that is wrong by 20 percent, the decision makes itself.
Common BI problems
Most BI problems show up as one of a few symptoms. The table maps each to the guide that diagnoses it.
| Symptom | Usual cause | Guide |
|---|---|---|
| The same metric differs by dashboard | No shared definition; each tool computes its own formula, window or filter | Metrics layer and definitions |
| Query totals exceed what the platforms report | Join fan-out, double counting across grains, missing distinct counts | SQL for marketers |
| Dashboards nobody trusts or owns | Sprawl, copies of copies, no certification or retirement | Dashboard governance |
| The team cannot answer joined or historical questions | Still on platform reporting; no warehouse and no history | Reporting vs BI |
| The raw export exists but nobody uses it | No SQL skill in the team; no modelled tables between export and dashboard | SQL for marketers, BigQuery export basics |
| Dashboards are slow or time out | Blending and calculating in the BI tool instead of modelling upstream | Looker Studio performance |
| Finance and marketing numbers do not tie | Different grain, time zone or counting unit on each side, undocumented | Metrics layer and definitions |
| A Looker Studio chart disagrees with the GA4 interface | Different dates, mixed scopes, thresholds, blends, averaged ratios or a stale cache | Looker Studio not matching GA4 |
| Queries on the raw export disagree with the GA4 interface | Modelled data, session logic in SQL, time zones, source scope or incomplete tables | BigQuery export not matching GA4 |
| Ad platform conversions in the warehouse fall short of the platform for recent days | Conversions credited after the load, with no refresh window for late arrivals | Loading ad platform data |
| A broken table reaches a dashboard before anyone notices | No tests on keys, required fields, mapping, freshness or volume | Data tests for marketing tables |
| A dashboard keeps showing old figures, or its latest days never arrive | A cache or extract, a stalled warehouse load, a fixed date range, provisional source days or an early delivery | Looker Studio report not updating |
| A metric moved and the first plausible story is taken as the answer | No method for ruling out data changes, separating mix from rate and finding the segment | Explaining a metric change |
| A comparison with last month or last year points the wrong way | Misaligned weekdays and holidays, partial periods, tracking breaks or seasonality | Comparing periods |
| Campaigns paused or rewarded on a handful of conversions | Rates built on small counts read as news | Reading small numbers |
| The analytics backlog fills with charts nobody acts on | Requests answered in arrival order, without the decision behind them | Analytics request process |
| An analysis reaches a decision maker checked only by the person who wrote it | No second-person review of definitions, joins, comparisons, the counts behind rates and causal wording | Reviewing an analysis |
| A wrong number is found after the report has gone out | A double-counting join, a dropped filter or an early export, fixed quietly instead of restated | Correcting a number |
For the consumption side of the same work, meaning the audiences a dashboard serves, the cadence it runs on and the narrative that goes with the numbers, the reporting and BI overview is the companion page to this section.
